Fix NeverMetered API errors: 401, 403, 404, 429 and more
Updated 10 November 2026
Most errors come from five causes: a wrong key, a wrong model name, an unverified email, too many requests at once, or a conversation that's outgrown the context window. Find your error below. The examples are the API's actual response formats; a few values that vary, such as your plan's numbers or your dashboard link, are shown as placeholders in angle brackets.
Check these first (30 seconds)
These two commands tell you whether your key works and which model name to use. Put your key in the environment variable first.
export NEVERMETERED_API_KEY="YOUR_API_KEY"curl https://api.nevermetered.com/v1/models \
-H "Authorization: Bearer $NEVERMETERED_API_KEY"If that lists qwen3.8-27b, your key and the network are fine and the problem is in your tool's settings.
The errors you're most likely to hit
| Status | Code | Usually means | Jump to |
|---|---|---|---|
| 401 | invalid_api_key | Missing, mistyped or revoked key | 401 |
| 403 | email_verification_required | You haven't clicked the verification link yet | 403 |
| 404 | model_not_found | Wrong model name | 404 |
| 429 | concurrency_limit_exceeded | More requests at once than your plan allows | 429 |
| 429 | Free requests used | Today's free allowance is used up | 429 free |
| 400 | context_length_exceeded | The conversation is longer than your plan's context window | 400 |
| 503 / 504 | request_queue_timeout, service_unavailable | Busy or temporarily out of capacity: retry | 503 |
Less common responses, and what to do about each:
- 413
request_too_large: the request body is over the size limit. Send less. - 502
upstream_error: the model server failed to respond. Retry shortly. - 429
rate_limit_exceeded: too many requests too fast. Wait about 10 seconds. - 403
account_suspendedorfree_tier_unavailable: the account can't use the API right now. Contact support. - 504
max_stream_duration: one generation hit your plan's time cap. Ask for a shorter answer, because retrying the same request won't help. - 503
request_cancelled: the service cancelled the request. Retry.
401 invalid_api_key
{
"error": {
"message": "Invalid or missing API key.",
"type": "authentication_error",
"param": null,
"code": "invalid_api_key"
}
}The API didn't recognise the key. Look for a missing header, a stray space or quote copied with the key, a key you revoked, or an old key from before you made a new one. The key starts with nfk_. Keys are shown once, so if you can't find it, make a new one. Tools that use the Anthropic format send the key as x-api-key or a bearer token; both work.
403 email_verification_required
{
"error": {
"message": "Verify your email address to use the free requests (<your dashboard link>), or upgrade to a paid plan.",
"type": "permission_error",
"param": null,
"code": "email_verification_required"
}
}Free requests need a verified email. Click the link in the verification email (press Resend email on your dashboard if it didn't arrive, and check spam). It starts working the moment you're verified, with no new key needed. Curl prints this as "The requested URL returned error: 403" when run with --fail.
404 model_not_found
{
"error": {
"message": "The model 'gpt-5' does not exist or you do not have access to it. Use 'qwen3.8-27b'.",
"type": "not_found_error",
"param": "model",
"code": "model_not_found"
}
}Your tool asked for a model that doesn't exist here, usually its own default (a gpt-… or claude-… name). The only model is qwen3.8-27b, written with a hyphen. If you've run a model through other tools, note that some write it with a colon, like qwen3.8:27b: use the hyphen here. Claude Code needs several model variables set at once, which the setup command does for you.
429 concurrency_limit_exceeded
{
"error": {
"message": "Too many concurrent requests: your plan allows <N> running and <M> queued. (the message continues with an upgrade link)",
"type": "rate_limit_error",
"param": null,
"code": "concurrency_limit_exceeded"
}
}All your parallel streams are busy and the queue is full. It comes with Retry-After: 1, so a well-behaved tool retries in a second. If you see it often, lower your tool's parallelism (subagents, batch size, CI jobs), or put background work on a capped key. How parallel streams work shows how to read your peak and decide whether you need more. A key with its own cap gets a shorter message: "Too many concurrent requests for this API key."
429 when the free requests run out
You've used all <N> free requests for today. (the message continues with an upgrade link)Free requests reset at 00:00 UTC. Coding agents send one request per step, so a free allowance goes quickly. This one deliberately has no Retry-After header, so tools don't sit silently for hours. Paid plans have no daily request limit: see pricing.
400 context_length_exceeded
prompt is too long: <tokens> tokens > <limit> maximum. Your plan's context window is <size>; shorten the conversation.Your conversation, plus the answer being generated, is larger than your plan's context window. The Claude Code and OpenAI-format wordings differ, but the fix is the same. Let your tool compact the conversation (Claude Code does this on request or automatically) or start a fresh session. If Claude Code compacts too late, it's usually missing the context-size setting: see the context fix.
503 and 504: busy or out of capacity
- `503` / `504 request_queue_timeout`: The service stayed busy longer than the queue timeout and nothing was generated. Retry with backoff, and give the tool a generous request timeout.
- `503 service_unavailable`: No GPU capacity is online at the moment. It's temporary and your setup is fine: retry in a little while, and check the status page.
My answers stop mid-sentence. Is that an error?
length. Ask for shorter steps, or check your plan's output limit on Plan & billing.How do I know it's a NeverMetered error and not my tool's?
x-request-id header: quote it if you contact support.Where do I check whether the service itself is down?
- NeverMetered docs: curl and API quickstart (checked 10 November 2026)
- NeverMetered pricing and plan limits (checked 10 November 2026)